Marble Cognition's AI Coworkers Are Learning the Job by Watching the Screen

The most expensive part of any enterprise automation project is the upfront work: mapping the process, documenting the steps, and translating human judgment into brittle rules. Marble Cognition, a Munich-based startup founded in 2025, is betting it can skip that bill entirely. Its desktop application simply watches how employees work, recording clicks and keystrokes across business applications, and then builds AI agents to take over the repetitive parts [PERPLEXITY SONAR PRO BRIEF]. The company calls these agents "Marble Coworkers," and a $3.75 million pre-seed round from Heartcore and Seedcamp suggests investors see a path to productizing that observation [Tech Funding News, November 2026].

The Wedge of Silent Observation

Marble's approach is a direct challenge to the consultant-heavy, manual-discovery phase of robotic process automation. Instead of asking teams to document their workflows, the software observes them in situ, excluding private apps and tabs and anonymizing data before analysis [PERPLEXITY SONAR PRO BRIEF]. Employees can pause recording at any time. The output is a structured process map that becomes the training ground for an AI agent capable of executing those same steps within the existing software stack, with a human kept in the loop for judgment calls [PERPLEXITY SONAR PRO BRIEF]. For procurement teams, the initial appeal is clear: reduce the time and friction to identify automation candidates. The company reports one U.S. financial-services firm used its agent to cut case-handling time by 90%, though the customer was not named [Tech Funding News, November 2026].

Early Traction in a Niche European Market

The founding team, Felix Rösner and Christophe Schmit, has focused its early sales motion on a specific slice of the German and European business landscape. Their named initial customers,fitness tech company EGYM, private equity firm Mutares, and consultancy Enam,point to a deliberate strategy [Tech Funding News, November 2026]. These are organizations with complex, document-heavy back-office operations, often housed within private equity portfolios where operational efficiency is a direct lever on value. Co-founder Christophe Schmit's prior experience at Enam, where he reportedly deployed AI agents for large company operations, likely provided both a first customer and a template for the pitch [Seedcamp Job Board]. The company is now hiring its first non-technical growth role in Munich, with a posted salary range of €75,000-€110,000, signaling a move from pure product development to scaled sales [Wellfound].

Pre-Seed Round | 3.75 | M USD

The Realistic Competitive Set

Marble's ideal customer profile is not a tech-native startup with clean APIs, but a legacy-rich enterprise,think a private-equity-owned professional services firm or a large European industrial,where processes are tribal knowledge and software sprawl is the norm. For them, the alternative to Marble isn't another AI agent startup; it's the status quo of manual work or the high-cost, high-touch implementation of traditional RPA platforms. The company's bet is that its silent observation and low-friction setup will win where lengthy discovery workshops and integration projects have failed. The competitive moat, if it exists, will be built on the proprietary dataset of observed processes it accumulates, making its agents more context-aware for similar vertical workflows over time.

Where the Implementation Risks Lie

For all its promise, Marble's model introduces several execution risks that any serious buyer will scrutinize. The procurement cycle for this category is long, and the budget owner is often unclear,is this an IT spend, a department-level productivity tool, or a strategic operations initiative? Furthermore, the renewal motion is entirely unproven. A successful proof-of-concept that saves 250 hours a month, as the company claims in one case study, must translate into a multi-year enterprise contract at a price that justifies the inference costs of running persistent AI agents [YouTube]. The risks break down into three core areas:

• Data governance and privacy. While Marble excludes private apps and anonymizes data, convincing large, regulated European enterprises to install a persistent screen recorder requires overcoming significant security and compliance hurdles [PERPLEXITY SONAR PRO BRIEF].
• Process complexity. Not all work is repetitive, click-based labor. The product's effectiveness hinges on its ability to correctly interpret and model processes that involve unstructured communication or external variables.
• Economic model. The value is in time saved, but the cost is in cloud inference and engineering support. Marble must price its service at a point that captures significant value for the customer while maintaining healthy margins as it scales.

The $3.75 million in funding, closed in under two weeks according to Rösner, will be used to scale inference infrastructure, grow the Munich engineering team, and launch U.S. operations [Tech Funding News, November 2026]. The next twelve months will test whether the product can move beyond early-adopter consultancies and financial firms and land logos where the procurement process is measured in quarters, not weeks. For Marble Cognition, the screen is just the beginning; the real work is building a business process layer that enterprises are willing to pay for, year after year.

Sources

1. [PERPLEXITY SONAR PRO BRIEF] Marble Cognition product and approach summary
2. [Tech Funding News, November 2026] Can AI learn your job simply by watching you work? Munich’s Marble has raised $3.75M to find out | https://techfundingnews.com/marble-cognition-3-75m-ai-agents-workflow-automation/
3. [Seedcamp Job Board] Growth & Operations, First non-tech hire @ Marble Cognition | https://talent.seedcamp.com/companies/marble-cognition/jobs/87702246-growth-operations-first-non-tech-hire&gsessionid=_WpqHOsO5DWcVGyeDSeCpBhbc7y9jyixTxqZ16d-KXL_mmsqxCrLpg
4. [Wellfound] Startup Jobs & AI Recruiting Platform | https://www.angel.co/
5. [YouTube] How Marble’s AI Clones Uncovered 40% Redundant Work. And Saved 250+ Hours/Month | https://www.youtube.com/watch?v=sc-A0MPiZxE